Characterising and analysing performance sequences in rugby league using match events data

The ability to accurately evaluate player and team performances in professional sport is particularly valuable. Doing so provides competitive advantages include extracting important information regarding the tactical strategies of future oppositions and producing player rating systems. A common method of evaluating player and team performances is via expected possession value (EPV) models. EPV models assign a value to every location and/or action on the pitch, which reflects the probability of points being scored within a given time period. EPV models have been produced in several sports, including football, basketball and ice hockey. However, there is limited research surrounding these models in rugby league. Rugby league has a unique set of rules, including a six tackle attacking set and five possible scoring options at the end of a possession. These two factors, alongside the poor data availability in the sport ensure that the majority of previous methods cannot be adapted for use in rugby league. Therefore the aim of this thesis was to develop new methodologies evaluating player and team performances in rugby league. In the first section of this thesis (studies 1 and 2), previous Markov models using zonal approaches were applied, adapted and extended in rugby league to provide insights into player and team performances. Six EPV models were produced with varying zone sizes using Markov Reward Processes. The Kullback-Leibler Divergence was used to evaluate the zone sizes which could reproduce future team attacking performances. The model was then extended to incorporate actions and context nodes using Markov Decision Processes. Novel methods of evaluating player and team performances were also produced. In the second section (studies 3 and 4), novel models producing smooth pitch surfaces were developed. The spatial trends of team attacking performances were evaluated using Kernel Density Estimation. Two novel Wasserstein distance metrics were used to provide valuable insights into team performances. A novel approach to the estimation of individual possession outcomes was also proposed using a Bayesian mixture model approach. The model used linear and bilinear interpolation techniques for its weights to produce a smooth pitch surface. Novel performance metrics evaluating player and team performances were also created. The research provides new methodologies for use within rugby league, providing zonal and smooth EPV models through which player and team performances can be evaluated. Professional experts were impressed with the results they provided and validated their use within the sport.